AI & Machine Learning
On-premise AI/ML tools: multi-LLM token counter, RAG chunking visualizer, synthetic data generator, Ollama vs cloud cost optimizer.
All AI & Machine Learning Tools
LLM Cost Optimizer - TCO Multi-Provider
Calculate the monthly and annual TCO of an LLM workload across 8 cloud providers (GPT-5, Claude, Mistral, DeepSeek, Gemini) vs Ollama on-prem on Hetzner CCX13. Break-even analysis with compound growth. Client-side only.
Open tool →Synthetic Data Generator
Generate synthetic datasets from JSON Schema with built-in Italian Faker: names, addresses, codice fiscale, partita IVA. Export to JSON, CSV or SQL. 100% client-side, no LLM, no data sent to server.
Open tool →Multi-Model Token Counter
Count tokens and estimate USD cost for 9 LLM models (GPT-5, Claude, Llama 3, Mistral, DeepSeek, Gemini) side-by-side. Client-side, no install, multilingual. Estimate ±10% vs official tokenizer.
Open tool →RAG Chunking Visualizer
Compare 8 RAG chunking strategies side-by-side: fixed-size, sentence, paragraph, recursive, semantic-cosine, sliding-window, markdown-header, semantic-token-bucket. Jaccard similarity heatmap included. 100% client-side, zero data sent.
Open tool →Prompt Eval Harness - LLM A/B Test On-Prem
A/B test prompts for LLMs via Ollama on-prem. Evaluate up to 20 prompts × 3 models (LLaMA 3.1 8B, Qwen 2.5 7B, Mistral 7B) with a results matrix: latency, token count, binary keyword-match score. Privacy guaranteed - no data sent to the cloud. USP vs LangSmith/Helicone/Promptfoo: on-prem + multilingual + Italian + €0 cost.
Open tool →AI Agent Trace Viewer - Multi-Format Inspector
Multi-format AI agent trace inspector (LangChain, CrewAI, AutoGen, OpenAI Assistant). Upload trace JSON, tree visualization of steps (LLM call / tool call / observation / decision), per-node drill-down, latency/cost stats. 100% client-side, privacy-first.
Open tool →Embedding Visualizer 2D/3D - UMAP/t-SNE/PCA
Visualise precomputed embeddings or generate from text via on-prem Ollama (local Hetzner) with UMAP, t-SNE or PCA dimensionality reduction. Inline SVG 2D or Three.js 3D rendering. JSON/CSV upload, visual cluster detection, image export.
Open tool →AI System Prompt Analyzer - Score 0-100 + Rewrite
Analyse LLM system prompt quality based on Anthropic prompt engineering guide + OpenAI cookbook. 7 checks (role, format, examples, edge case, safety, length, clarity), weighted 0-100 score, severity-ranked issue list + suggested rewrite skeleton. Pure client-side.
Open tool →Vector DB Schema Designer - RAG cross-DB
Cross-vector-DB schema generator for RAG. Pick embedding model (12 options: MiniLM, MPNet, OpenAI ada-002, Cohere multilingual, Voyage, BGE etc.), chunking strategy, target vector DB (Redis Stack HNSW, Qdrant, Weaviate, Pinecone, Postgres pgvector, Milvus). Output: DDL/JSON schema + index params + Hetzner cost estimate.
Open tool →SLM Picker - Hetzner CCX + EU Languages
Filter open-weight Small Language Models (≤32B params) by Hetzner CCX size + EU languages + use case. 18 models updated 2026 (Llama-3.x, Mistral, Mixtral, Qwen2.5, Gemma-2, Phi-3, DeepSeek-V2, Aya, CroissantLLM). Ranked output with VRAM/RAM, t/s throughput per CCX, license, fine-tuning support, maturity score.
Open tool →Vibe Coding ROI Estimator - AI Dev Tools
ROI calculator for AI-assisted development (Cursor, Claude, GitHub Copilot, Codex). Input: team size, dev avg salary, task distribution (greenfield/maintenance/refactor/code-review/docs), tool subscription cost, expected velocity multiplier. Output: yearly ROI + payback period + saved hours by task type. KB cluster Marcello Ascani S5 AI engineering.
Open tool →LLM Cost per Output Calculator - 12 Models 2026
Per-output type cost calculator (chat reply, RAG, summarization, code-gen, agent-loop, embeddings, classification). 12+ models updated 2026 (GPT-5, Claude Sonnet 4.6, Gemini 2.5, DeepSeek-V3, Qwen2.5, Mistral Large 2, Ollama llama3.1:8b on-prem €0). Cost/day + cost/month + unit cost + provider comparison. Series 33 FinOps AI.
Open tool →LoRA Adapter Config - Fine-Tuning HF PEFT
LoRA/QLoRA fine-tuning configuration lookup for LLMs. Input: base model (Llama-3.1-8B/70B, Mistral-7B, Qwen2.5-14B, Gemma-2-9B), task type (instruction-tuning, domain adaptation, code, reasoning, multilingual), VRAM, dataset size, quantization (fp16/qlora-4bit/qlora-8bit). Output: r, alpha, target modules, learning rate, batch size, ready-to-use Hugging Face PEFT JSON config. KB Karpathy LoRA paper.
Open tool →LLM Faithfulness Checker - RAG Groundedness
Verifies LLM output faithfulness vs context (RAG groundedness). Jaccard overlap heuristic + numeric contradiction detection + unanchored key terms. 0-100 score, sentence-level breakdown (supported/unsupported/contradicted), evidence spans. RAGAs, TruLens, DeepEval frameworks.
Open tool →